Ara — Global Decentralized Infrastructurefor Payments, User Identity, Hosting, Streaming,

and Ownership of Digital Files and Content (Draft)

Eric Jiang, Charles Kelly, Joseph Werle, Tony Mugavero, Vanessa Kincaid

Last Updated February 28, 2020 (Partial**)

Abstract

The internet has become a shadow of its former self, dominated by a few largecompanies that exercise total authority over all the information by controllinghow it flows around the world, how much it costs, and how and when it canbe consumed. Ara is a decentralized platform and suite of protocols designed tofix that. Through licensing and selling digital files and content using a novelProof-of-Ownership system, Ara handles global data delivery and supports purchasingof those assets with the native Ara token. In doing all this, the Ara platformalso utilizes a ubiquitous and distributed user ID and wallet system which allowsusers to retain ownership of their personal information. In effect, Ara is a newmodern mental model around how information on the internet is hosted anddelivered and how consumers use and pay for it; it brings about a new paradigmnot only for businesses but also for consumers in that they can contribute in thesystem to earn rewards by hosting and participating in the network. Peer-to-peer (P2P) file sharing, blockchain technology for ownership and licensing, anddistributed computing are all combined into a single efficient and decentralized system.

Numerous constituents benefit from Ara. Consumers can use their idle stor-age, bandwidth, and processing power to earn Ara tokens–akin to Airbnb forcomputing–and they can use those tokens to purchase content. Businesses save ondelivering this information to people through P2P technology, which in turn lowerscosts to consumers and other businesses. Anyone can participate as a data center inthe network to earn rewards. Digital creators, game and software developers, movieand TV studios, and publishers use Ara tokens to publish licensed content into thenetwork, earning more on their work by generating larger revenue shares and givingthe rest to their fans for the hosting costs. It’s a win-win-win. Consumers arerewarded, publishers earn more, and businesses improve the bottom line. This is alldone in a decentralized and net neutral way, such that no intermediary companiescan throttle a single business delivering information and content to you.

** This is a partial update to the whitepaper from June 2018. A more thorough update will bereleased in the coming months.

Note: Ara is under active research and development. This paper is subject to change. Thelatest version will be available at https://ara.one. Please direct any comments and suggestions tohello@ara.one.

Contents1. Introduction 2

1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.2 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.3 Platform Services . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2. Platform Overview 42.1 AraID . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.1.1 Decentralized Identity . . . . . . . . . . . . . . . . . . . . . . . 42.2 Decentralized Content Delivery Network (DCDN) . . . . . . . . . . . 5

2.2.1 Ara File System (AFS) . . . . . . . . . . . . . . . . . . . . . . 52.2.2 Token Usage . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.3 Ara Protocol Suite . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.3.1 Rewards and Incentives . . . . . . . . . . . . . . . . . . . . . . 72.3.2 File Delivery . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82.3.3 Smart Contracts . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3. Future Development 103.1 Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103.2 Ara Name System (ANS) . . . . . . . . . . . . . . . . . . . . . . . . . 10

4. Acknowledgements 10

Acronyms 11

References 11

Appendices 13

I. Mature Ara Platform Token Economics (Draft) 13Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13The Ara Token . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14Market Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14Incentive Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14Network Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

II. DCDN Cost Analysis by Lester Kim 17Uploader’s Profit Maximization . . . . . . . . . . . . . . . . . . . . . . . . 17Distributor’s Cost Minimization . . . . . . . . . . . . . . . . . . . . . . . . 19Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

III. Expected Ara Rewards Analysis by Lester Kim 24Network Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

1

1. Introduction

1.1 Background

The hypermedia landscape today is outdated. Aggregators and app stores arestrengthening their grip on content creators; traditional content delivery networksare inefficient and expensive; cloud compute is centralized to a select few gatekeepers;and data is stored not by those who own it, but by those who profit from it. Contentpublishers and creators are forced to inflate prices, offloading the costs for thisslow and expensive system to consumers and resulting in lost value for publishers,consumers, and creators alike.

With video poised to comprise over 80% of all internet traffic by 2021 [3], the costof content has continued to rise as file sizes and costs of delivering that content goup. 4K, VR, and AAA games are all contributors to this trend. Consumers arenot only paying more for transactional content and subscriptions, but they have todeal with a complex and abusive system of advertising to view free content. Thesefactors exacerbate the problem of piracy, leading to billions of dollars in losses tocontent owners [16][12][4], and they result in the introduction of tools to skip orremove advertising such as ad-blockers. This in turn creates ad-blocker-blockers andmore expensive subscription services as consumers avoid ads and content owners tryto recapture lost revenue. It’s a vicious cycle.

Peer-to-peer (P2P) file distribution architecture emerged as a response to theseinefficiencies, evolving from hybrid solutions that incorporated centralized serverssuch as Napster to completely decentralized solutions such as Gnutella and eventuallyBitTorrent. Today, P2P file delivery is so cost-efficient that companies such asMicrosoft use it—not their own Azure infrastructure—to save on Windows 10distribution costs.

However, while cost-efficient, P2P file sharing networks have been historically rifewith free-riding, piracy, hacking, and black markets when used in public environments.There was no trust that the individual uploading content had the right to do so, andno way to verify that the content seeded was delivered as the content owner intended.There was also no reward for storing and sharing the content, so users were notincentivized to remain seeds in the system of delivery. Peers would leech contentfor themselves and not continue to participate in seeding that content out to otherpeers. To combat this, P2P architectures began to incorporate incentive mechanisms,such as barter strategies, reputation systems, and proprietary currencies. However,even these mechanisms have their share of problems and are subject to Sybil andwhitewashing attacks.

1.2 Overview

In this white paper, we present Ara: a community-driven decentralized and dis-tributed compute and content delivery platform. Ara enables any device in the worldto become part of a global supercomputer, database, and delivery network all atonce by utilizing its unused processing, storage, and bandwidth capacity. Together,

2

these devices form the Ara network, an ecosystem in which anyone can participateand benefit.

Fundamentally, the network is comprised of an overlapping community of consumers,service requesters, service providers, and software developers, each with their ownincentives for adoption. With Ara, service requesters can have at their fingertipsa vast reserve of compute resources along with an ever-expanding library of dis-tributed services. Service providers, who have already paid for their devices—be ita smartphone, laptop, or gaming console—can begin making a return renting outunused resources. The only requirement to begin earning rewards is to submit asmall deposit that acts as a hold on the account. This hold can be withdrawn at anytime, but it is required to earn and redeem any rewards. Software developers canleverage the unparalleled scale of Ara’s ecosystem to accomplish hefty compute tasksand create novel distributed services for requesters and providers to participate in.Meanwhile, consumers can go about their daily lives as they would all while beingrewarded for watching the shows and listening to the music they love.

Thus, anyone with spare compute resources can immediately act as a service fulfillerto earn rewards for helping distribute content, while anyone looking for remoteresources can request Ara decentralized services and enjoy enhanced security, fileavailability, and delivery speeds for a fraction of the cost compared to traditionalcloud compute service providers. Because Ara removes the burden of purchasing andmanaging infrastructure, content creators of all kinds stand to benefit—from theindie artist who can now freely self-publish his new album without going through arecord label, to the large media conglomerate who no longer needs to go throughaggregators to reach its audience. Ara relies on the resources provided by members ofthe network; the larger the network grows, the more robust and efficient it becomes.

1.3 Platform Services

The Ara platform is comprised of three (3) core services and systems:

1. AraID: AraID establishes secure, decentralized, and verifiable global identitiesfor all agents and content on the Ara platform, giving control of data back totheir rightful owners.

2. Decentralized Content Delivery Network (DCDN): DCDN serves asAra’s network of underlying peer-to-peer, secure distributed file systems andstorage networks (AFSs) which support content integrity, incentives, versioning,and decentralized identities.

3. Protocol Suite: Ara is connected through a secure suite of protocols whichenable trustless interoperability between DCDN, AraID, and the EthereumBlockchain.

3

2. Platform Overview

The Conflict-Free File System Network (CFS-Net) is the backbone of Ara’s peer-to-peer dis-tributed file system, AraID, and DecentralizedContent Delivery Network (DCDN). Leverag-ing an underlying Merkle tree structure andthe Syncable Ledger of Exact Events Protocol(SLEEP) [1] file format, CFSNet addresses manyconcerns with traditional file transport—bothclient-server and P2P—and improves upon ex-isting technologies such as IPFS by providingcryptographically-ensured content integrity aswell as versioning and revision history. The net-work is made up of isolated file systems calledCFSs. Furthermore, each CFS instance imple-ments a subset of the Filesystem Hierarchy Stan-dard (FHS) [8], supporting partitions and allow-ing each directory to exist as a self-containedCFS archive with its own access levels. Of thesepartitions, AFS uses the /home and /etc par-titions to store AFS content and metadata, re-spectively. Each CFS partition is publicly iden-tifiable throughout the network using a uniqueEd25519 32-byte public key generated at thetime of creation. A CFS’s public key grantsread-only access to the file system, where onlythe holder of the private key can update andpublish the contents contained within.

2.1 AraID

AraID is responsible for creating and resolvingsecure and verifiable decentralized representa-tions for all users and content on the Ara Plat-form. Fully compliant with the W3C (Decentral-ized Identifier (DID)) spec [15], AraID uses DIDDescriptor Objects, or DDOs for short, to repre-sent users and content (see Figure 1 ). DDOs aresimple JSON-LD documents which define meth-ods for authentication and authorization as wellas other identity attributes, including serviceendpoints and private communication channelscontrolled by the owner [15]. Because DDOsnever store Personally-Identifiable Information(PII) [15], these service endpoints and commu-nication channels identify secure means of ob-taining it and thus allow entities self-sovereignty

over their private data and online identities.

2.1.1 Decentralized Identity

For all users and content on the Ara platform,an AraID is generated in the form of:

• did:ara:ee93189c629cdaf949fd57bac5b005b916936d2a5c680640fd1aedc8315730a0

AraID implements a Universal Resolver method,denoted by the second component of the DID(ara above) as part of the Decentralized Iden-tity Foundation system [11]. The method, alsoknown as a driver, defines how DIDs and DDOsare resolved within the Ara platform. Unlikeinternet URIs, DIDs do not require a centralauthority for registration or control and forma bijective correspondence with DDOs ratherthan the non-injective, non-surjective relation-ship found in TCP/IP and DNS.

The crux of AraID security is maintained cryp-tographically using Decentralized Public Key In-frastructure (DPKI)[13], wherein Ed25519 pub-lic keys are used as both the id portion of theDID (ee9318... above) and the public key of theCFS in which the corresponding DDO is stored.These documents contain a publicKey prop-erty which holds various keys used for digitalsignatures, encryption, and other cryptographicoperations. When an identity is created, this ar-ray is populated with the owning identity’s key,as well as the corresponding Ethereum account’spublic key.

For AFS AraIDs, the public key of the /etc par-tition containing the associated content meta-data is also stored. Since the key for this isstored in the AFS DDO, it can be resolved byany requester who has the AFS DID.

Whenever a new identity is generated, amnemonic phrase is used to seed the keypair.The mnemonic is returned to the owner for safe-keeping and for ease of maintaining the privatekey, allowing entities to easily validate own-ership of DIDs and restore accounts withoutneeding the private key.

4

Identity Archiving and Resolution

When an identity is created, it is initially writ-ten locally so that any local resolution can checkthe cache prior to falling back on the network.However, before an identity can be resolvedremotely, it must first be archived. Ara runsarchival nodes whose job it is to store theseidentities for future resolution.

Similarly to the archiver nodes, Ara also runsresolver nodes responsible for querying thearchivers for requested DDOs. Resolver requestsfirst look to locally resolve identities that maybe stored on disk before reaching out to thenetwork for remote archivers that have archivedthe AraID in question.

{'ddo': {

'@context': 'https://w3id.org/did/v1','id': 'did:ara:ee9318...','authentication;: [{

'type': 'Ed25519SignatureAuthentication2018','publicKey': 'did:ara:ee9318...#owner'

}],'publicKey': [{

'id': 'did:ara:ee9318...#eth','type': 'Secp256k1VerificationKey2018','owner': 'did:ara:ee9318...,'publicKeyBase58': 'H3C2AVvLMv6gmMNam...'

}],'service': {

'ens': 'https://etherscan.io/enslookup',}...

}}

Figure 1: Example DDO

Ethereum Account

Each identity is created with an Ethereum ac-count and an associated Ethereum wallet, re-coverable using a generated random mnemonicduring identity creation. Since the Ethereumaccount and the identity itself are determin-istically created using this mnemonic, a usercan recover their full identity, including theirEthereum account and wallet using just thismnemonic.

AraID is designed to support any accountbacked by public key cryptography. Thus, it isagnostic to the types of cryptocurrency accounts

it can support, and it can easily be associatedwith cryptocurrency wallets of any kind.

2.2 Decentralized Content DeliveryNetwork (DCDN)

DCDN is Ara’s solution to scalable, decentral-ized hypermedia and digital asset distribution.At it’s core, DCDN is composed of a networkof Ara File Systems (AFSs), CFS implementa-tions which house content and their associatedmetadata.

2.2.1 Ara File System (AFS)

AFS is a flavor of CFS tailored to meet the spe-cific needs and goals of Ara. AFS leverages twoexisting partitions that CFS implements, the/home and /etc partitions. These partitions,implemented as a subset of the Filesystem Hi-erarchy Standard (FHS) [8], are responsible forthe raw binary data and the content metadata,respectively. The /home partition can only beaccessed after a user has purchased or beengranted access to the AFS’s content, whereas the/etc partition containing the metadata can beaccessed regardless of content ownership. TheAFS owner may define a schema for the meta-data so that it can be parsed by a requester. Theprotocol does not enforce a strict standard asto how metadata should be structured, but werecommend Schema.org as a reference to exist-ing paradigms to best support interoperabilitybetween decentralized services.

When an AFS is initially created, an AraIDis created using a BIP39 [5] random 12-wordmnemonic phrase. The generated DID is usedas the public key of the AFS, and the corre-sponding DDO’s authentication property isamended to include the owner’s DID. By doingso, an AFS’s owner can be determined fromresolving its DID.

An AFS is created for each piece of contentintroduced into the system. This can be a sin-gle file such a movie, or a collection of filessuch as a game. To cryptographically verify

5

content ownership, two sets of byte buffers arewritten to the Ethereum blockchain. The firstare the metadata.tree entries, which representthe serialized merkle tree of the data containedwithin the data storage layer. The second isthe metadata.signatures file, containing sig-natures of the serialized tree’s root nodes.

2.2.2 Token Usage

Initially, owning Ara tokens with respect toDCDN grants the following capabilities:

1. The ability to purchase and download con-tent within the network.

2. The ability to participate in P2P file de-livery and earn rewards on any content bysubmitting an Ara deposit that acts as ahold.

6

2.3 Ara Protocol Suite

The following subsections define the core protocols of the platform, describe each part of thesystem in detail, and explain the interoperability between them.

Figure 2: Illustration of the Ara Protocol

2.3.1 Rewards and Incentives

Traditional peer-to-peer file sharing systems,such as BitTorrent, rely on altruistic behaviorand lack effective incentives [7] for peers to up-load as much to the network as they download,creating an imbalance where leechers (usersdownloading a distributed file) can easily domi-nate a swarm (all peers downloading or upload-ing a distributed file). This imbalance, a formof The Free Rider Problem [6], is undesirable ina healthy network since leechers often benefitfrom the network at the expense of others with-out offering anything in return. Ara implementsa reward system as an incentive mechanism to

mitigate this network inefficiency and increasefile availability by applying a small cost to eachdownload and distributing that cost as a rewardto every peer who seeds the download. Depend-ing on the payment model, this cost could bebaked into the total cost of the content as areward allocation; for "free" content, this costcan replace traditional ads or data collection(how users pay for "free" content today). Sinceeach download becomes a source of rewards,over time, network participants stand to earnmuch more from rewards then they pay up frontfor the download.

7

2.3.2 File Delivery

File Delivery is the main mechanism for up-holding Ara’s content delivery network and forparticipants to earn rewards. Ara’s file deliveryprotocol begins with a four-step handshake.

1. Alice, the content requester, broadcastsa download request for a piece of contentover some network discovery protocol (CF-SNet implements several strategies, includ-ing mDNS and BitTorrent).

2. Bob, a license validator and content deliv-erer, receives the broadcast and respondswith his file availability.

3. Alice selects peers from the pool (theswarm) of responses and sends a messagewith her intermediate public key alongwith her DID.

4. Bob cryptographically verifies Alice’s mes-sage and that she has purchased a licenseto the underlying AFS.

Once the handshake is complete, the file transferbegins.

2.3.3 Smart Contracts

Ara is initially launching on Ethereum mainnet,where smart contracts will serve as a centralcomponent of Ara’s protocol suite. These smartcontracts mediate and facilitate interoperabil-ity between DCDN, AraID, and the applicationlayer, ensuring that all non-transient propertiesand entities on the platform are registered onthe Ethereum blockchain, including:

• Published Content• Purchases• Rewards• Ara Balance

Ara’s smart contract architecture was designedwith security and modifiability in mind. In orderto support evolving conceptions of how AFSsshould be sold and purchased, how rewardsshould be handled and distributed, and how pay-ments should be processed and routed withinthe system, Ara deploys a Proxy Contractfor each published AFS. Proxies are deployed

through a Registry Contract, where they areassociated with a specific version of an AFSStandard, which defines the business logicfor AFSs. Whereas deploying the entire AFSStandard for each AFS would be costly anddifficult to update – the AFS would in essencebe stuck with a specific standard – proxy archi-tecture allows a single Proxy to be deployed forthe lifetime of an AFS and upgraded by chang-ing which AFS Standard version it references.Proxy architecture also ensures that only regis-tered Proxy addresses (i.e., valid AFS content)can be added to user libraries in the LibraryContract.

AFS Standard

The AFS Standard enables AFSs to have a de-fined, structured, and self-contained presence onthe Ethereum blockchain. It includes methodsfor purchase and rewards, as well as methodsfor storing the tree and signatures files fromthe metadata SLEEP register. The metadataSLEEP register stores metadata about the con-tent within an AFS, including filenames, sizes,and permissions, while the content SLEEP reg-ister stores the raw binary contents of the files.Within the metadata register, the tree file rep-resents the serialized Merkle tree that makesup the data in the content register, and thesignatures file stores the signed roots of theserialized tree. Whereas with CFS, these fileswould be stored on and read from disc, withAFS, they are written to and read from theEthereum blockchain. Since AFSs communicatewith this standard through their own Proxy,many different AFS Standards can coexist, al-lowing content creators to choose the standardthat best suits their needs.

The use of Proxies separates logic from stor-age, where the AFS Standard serves as the logiclayer for any AFS using that version of thestandard, and each Proxy serves as the stor-age layer for a single AFS. At a minimum, AFSStandards must implement an AFS Standardabstract class which enforces the implementa-tion of pricing, purchase, rewards, and storagefunctions.

8

For the most basic (and default) AFS Standard,prices can only be modified by the owner of anAFS. Upon purchase, this price is transferredfrom the purchaser’s Ara wallet to the owner’sAra wallet. The base AFS Standard enforcessupport for reward budgets, which must be sub-mitted prior to download. Once the downloadis complete, the budget is allocated betweenparticipating peers, who can then redeem theirrewards granted they have not withdrawn thebalance hold required to do so.

At the discretion of content creators, AFSStandards can also support a multitude of cus-tomizable commerce controls:

1. Royalties: Purchases can be customizedto distribute proceeds amongst many dif-ferent Ara accounts by percent break-down.

2. Bulk Purchases: Prices can be tieredbased on quantity purchased.

3. Resale Conditions: Purchased contentcan be resold a number of times for atleast a minimum resale price as specifiedby the content creator.

4. Ownership Transfers: The owner of anAFS can readily transfer ownership to an-other Ethereum address.

5. Pre-Orders: Content can be purchasedbefore it is available for download. Pur-chasers can submit a reward budget aheadof time so that they can begin download-ing an AFS as soon as it is available.

6. Scarcity: Content creators can define amaximum number of sales for an AFS, af-ter which the AFS becomes unlisted andunavailable for purchase.

These commerce controls provide content cre-ators of all kinds with the power to define theirown business and revenue models tailored fortheir needs to their exact specifications. Thecontrols can be combined to form interestingnew models which would be difficult to imple-

ment via traditional means. For example, some-one who enjoys remixing music can purchase thetracks from the original artist with defined resaleconditions and minimum resale prices, remix thesongs, and sell the remixed versions while stillpaying out the original artist. Whereas previ-ously combining these types of controls wouldrequire exorbitant amounts of time, capital, andlegal involvement – serving as significant barri-ers to entry for smaller content creators – theyare now available to everyone for free.

Registry

As part of the proxy architecture, the RegistryContract serves two main functions:

1. It serves as a Proxy factory

2. It tracks all AFS Standard versions

When an AFS is first published, the Registrydeploys a Proxy for that AFS and establishesa relationship between it and a specified AFSStandard. The Proxy consults the Registryfor the address of its respective AFS Standardwhenever it is called and delegates the call tothat address, where it is processed and returnedto the Proxy.

Library

The Ara network leverages the LibraryContract to create a canonical source of truthfor AFSs that a user has access to, whether pur-chased or otherwise. When content is purchased,the purchase function in the AFS Standardautomatically adds the AFSs DID to the pur-chaser’s library in the Library Contract. Thisallows any service that requires informationabout a user’s library to query the contractfor that information. The AFS DID that isstored on the blockchain allows any service toresolve to the underlying content. The Libraryenforces that only registered Proxies can addtheir corresponding AFS to a user’s library, en-suring that no one can tamper with anotheruser’s library without their consent.

9

3. Future Development

3.1 Modules

The Ara platform is fundamentally agnostic tothe types of distributed services that can run ontop of it. Modules are distributed and/or decen-tralized services which implement the ModulesAPIs and can be used interchangeably withinthe platform. Similar to the way the ERC-20Token Standard allows any token on Ethereumto be re-used by other applications [14], theModules APIs allow all distributed services im-plementing the interface to be used throughoutthe platform. This facilitates the formation ofa developer community dedicated to buildingan ecosystem of distributed services that eachhave the ability to utilize Ara’s rewards, pur-chases, and payments systems, in essence form-ing a rewardable distributed services economy.Modules that seek to utilize the rewards systemare expected to implement an additional smartcontract API in which they define their ownrewarding mechanism and methodology. EachModule smart contract stores its own respectivereward allocation accumulated through use ofthe distributed service and distributes it accord-ingly.

3.2 Ara Name System (ANS)

ANS, much like the Domain Name System(DNS) [9], is a decentralized way to register,query, invoke, or revoke certificates within theAra network. While DNS sits as part of the ap-plication layer of the internet—resolving human-readable URIs to underlying IP addresses sothat a user agent (e.g., web browser) can ob-tain and render the requested content—ANSresolves human-readable names to DIDs for ulti-mate resolution to DDOs. ANS uses the identityarchiver and resolver under the hood for the sec-

ond phase of resolution to provide DDOs fromDIDs. ANS essentially is both an archiver andresolver for human-readable URIs. An exampleapplication of ANS could be providing DDOsfrom hostnames, in the context of a web browserbuilt on top of Ara.

To discern between the various record types,each record stores a TYPE resource record, sim-ilar to how DNS classifies its own records [17].The TYPE field is represented by a numeric value,which allows other types of records to be storedin ANS in the future. The following table de-scribes the record TYPE:

TYPE Value DescriptionUSR 00 UserPCT 01 Published Content

Each supernode hub that comprises ANS runsa HyperDB instance [2]. A distributed, highly-scalable database like HyperDB provides severalfeatures that make it suitable for a system likeANS. The first is HyperDB’s use of tries: searchtrees where each node is a prefix to its childnodes. By storing names with tries, we canguarantee that even with thousands of entriesin the database, lookups are inexpensive andquick. Lookups in tries are O(n) where n isthe length of the key being searched. HyperDBalso makes use of vector clocks, which track thecausality of events within a distributed systemto prevent cases where nodes become desynchro-nized [10].

4. Acknowledgements

Special thanks to Kevin Faaborg, AndrewGrathwohl, and Brandon Plaster for their con-tributions.

10

Acronyms

AFS Ara File System. 3–5, 9

ANS Ara Name System. 10

CFS Conflict-Free File System. 4, 5

CFSNet Conflict-Free File System Network. 4

DCDN Decentralized Content Delivery Network. 3–5

DDO DID Descriptor Object. 4, 5, 10

DID Decentralized Identifier. 4, 5, 9, 10

DNS Domain Name System. 4, 10

DPKI Decentralized Public Key Infrastructure. 4

FHS Filesystem Hierarchy Standard. 4, 5

PII Personally-Identifiable Information. 4

SLEEP Syncable Ledger of Exact Events Protocol. 4

References

[1] Code for Science Buus, Ogden. Sleep - syncable ledger of exact events protocol. https://github.com/datproject/docs/blob/master/papers/sleep.pdf, Aug 2017.

[2] Mathias Buus. Hyperdb. https://github.com/mafintosh/hyperdb, Aug 2017.

[3] Cisco. Cisco visual networking index predicts global annual ip traffic to exceed threezettabytes by 2021. https://newsroom.cisco.com/press-release-content?type=webcontent&articleId=1853168, Jun 2017.

[4] Stewart Clarke. Piracy set to cost streaming players more than $50 billion, study says. http://variety.com/2017/tv/news/piracy-cost-streaming-players-over-50-billion-1202602184/, Oct2017.

[5] Palatinus et al. Mnemonic code for generating deterministic keys. https://github.com/bitcoin/bips/blob/master/bip-0039.mediawiki, Sept 2013.

[6] Russell Hardin. The free rider problem. https://plato.stanford.edu/entries/free-rider, May2003.

[7] Ahamad Jun. Incentives in bittorrent induce free riding. https://disco.ethz.ch/courses/ws0506/seminar/papers/freeriding_incentives.pdf, Aug 2005.

[8] The Linux Foundation LSB Workgroup. Filesystem hierarchy standard. https://refspecs.linuxfoundation.org/FHS_3.0/fhs-3.0.pdf, Mar 2015.

11

[9] Paul Mockapetris. Domain names - implementation and specification. https://tools.ietf.org/html/rfc1035, Nov 1987.

[10] Multiple. Vector clock. https://en.wikipedia.org/wiki/Vector_clock.

[11] Markus Sabadello. A universal resolver for self-sovereign identifiers. https://medium.com/decentralized-identity/a-universal-resolver-for-self-sovereign-identifiers-48e6b4a5cc3c, Nov2017. Accessed on 2018-04-24.

[12] Stephen E. Siwek. The true cost of sound recording piracy in the us economy. https://www.riaa.com/wp-content/uploads/2015/09/20120515_SoundRecordingPiracy.pdf, Aug2007.

[13] Rebooting the Web-of Trust. Decentralized public key infrastructure. http://www.weboftrust.info/downloads/dpki.pdf, Dec 2015. Accessed on 2018-04-19.

[14] Buterin Vogelsteller. Erc-20 token standard. https://github.com/ethereum/EIPs/blob/master/EIPS/eip-20.md, Nov 2015.

[15] W3C. Decentralized identifiers. https://w3c-ccg.github.io/did-spec, Apr 2018. Accessed on2018-05-01.

[16] Music Business Worldwide. Why does the riaa hate torrent sites so much? https://www.musicbusinessworldwide.com/why-does-the-riaa-hate-torrent-sites-so-much/, Dec 2014.

[17] Inc ZyTrax. Dns resource records (rrs). http://www.zytrax.com/books/dns/ch8/, Oct 2015.

12

Appendices

I. Mature Ara Platform Token Economics (Draft)

Overview

Traditional cloud services have risen to prominence due to the flexibility, agility,and cost savings they afford over purchasing and managing in-house infrastructure.Many cloud services use notoriously complex and obscure pricing models that involvelong-term commitments and contracts—often individually negotiated per customer—directly undermining the flexibility and agility they were meant to provide in the firstplace [1]. In recent years, P2P CDNs have begun to gain traction, with emergingSaaS’s touting hybrid solutions for video delivery. While these SaaS’s established ahighly scalable solution with a simpler pricing model by leveraging the machines ofvideo viewers, they maintain several sources of centralization, namely through theirbusiness model and their attribution of users as “second class citizens”, wherein theutilization of their machines occurs without their consent or financial compensation.Ara brings it one step further by rewarding network participants for the utilizationof their machines.

In order to supplant existing classical cloud infrastructure costs, the Ara platformimplements a native protocol utility token: the Ara token. This token can be usedon the Ara platform to create cryptoeconomic incentives for healthy and honestnetwork behavior, to enable more direct engagement between content consumersand creators, as well as to encourage adoption of the platform. The Ara token canbe seen as an encapsulation of the value members of the network provide, with eachtoken rewarded representing a marginal increase in network utility.

The Ara Token

Distributed services built using Ara’s SDKs, known as Modules, can be bought, sold,requested, and fulfilled all on the Ara network. Module tasks are outsourced topeers in the network who, upon successfully completing them, can be compensatedwith Ara tokens. In order to promote an open and competitive market, Araallows service requesters to define reward allocations, or bounties, for services theyrequest and service providers to define a minimum bounty to accept. Similarto Amazon Mechanical Turk, a marketplace for crowdsourcing tasks that requirehuman intelligence, Ara creates a marketplace for outsourcing distributed computeor networking tasks. Modules can be one-off on-demand tasks, such as distributedtranscoding, or they can be continuously recurring services, such as a P2P multiplayergame server.

13

Function

The Ara token can be used throughout the network in a variety of ways.

• For consumers, Ara tokens can be used to make any sort of purchase, rangingfrom digital content for enjoyment to new Modules to participate in

• Service requesters can use Ara tokens to initiate job requests and to set bountiesfor the successful completion of those jobs

• Service providers can deposit Ara tokens as a commitment to fulfill a task inreturn for rewards (the deposit can be returned upon request, but the providerwill no longer be able to earn rewards)

• Developers can use Ara tokens to deploy new Modules into the network

Since each of these roles heavily overlap, a service provider can use the same Aratokens earned as rewards through fulfilling a task to purchase a new piece of content,just as a developer can use Ara tokens earned through Module purchases to initiatea new job request.

Market Dynamics

Because members of the network have full agency in deciding which tasks or servicesthey participate in, the network forms a free-market in which economic equilibriumsemerge. Each Module will likely have its own economics governed by the behavior ofthe requesters and providers for that specific Module’s jobs. For example, distributedvideo transcodes might be a characteristically urgent job for video producers, resultingin price inelasticity of demand for distributed video transcodes (i.e., video producersare relatively indifferent to how much they might need to reward the service providers).Thus, a seller’s market forms in which distributed transcode service providers havethe advantage in determining reward allocation, driving the bounty up. Similarly, aP2P game server Module may be highly requested but have relatively few serviceproviders. Again, a seller’s market forms and the bounty is driven up. On theother hand, a distributed machine learning Module might be requested by fewpeople but have many eligible service providers. Due to the relatively low demand,many providers will miss out on opportunities if they set their minimum bountyrequirement too high. A buyer’s market forms, and the bounty is driven down.

It is important to note that it is not a requirement for Modules to set bounties,and there is no standardized model for how bounties should be setup. The goalsare to align the incentives of service requesters and service providers, to supportall types of incentive models, as well as to accommodate the varying fixed costs forinfrastructure and networking capabilities across the world.

Incentive Structure

To better understand how this model supports an alignment of incentives betweenservice requesters and service providers, we derive a general description of theeconomic interests of both parties. Service requesters want their requests to befulfilled for the lowest cost, while service providers want to optimize for the greatestnumber of highest paying services. In other words, it is in the best interest of both

14

requesters and providers to maximize network utility as long as both agree on aprice. Thus, the services that best balance bounty and work are most likely to befulfilled, creating market pressure that promotes both competitive bounties andinnovation in distributed service design to improve efficiency.

The variety of distributed services that can run on the platform necessitates flexibilityin determining unit-of-work-rewarded (UWR, the base unit of provable work bywhich bounties are split up and rewarded) and bounty model (the conditions thatgovern how bounty is paid). A P2P multiplayer game server might use number ofrequests served as its UWR, and a game developer invoking that server Modulemight decide that a subscription-based, recurring bounty model makes the mostsense. On the other hand, a distributed transcoding service might use number ofbytes transcoded per minute as its UWR, and a video producer might payout abounty per transcode.

Service providers can stake Ara tokens in order to participate and earn rewards.Like with bounties, service requesters can determine a minimum stake, if any, thatproviders must deposit in order to engage in the service. The minimum stakevalue should be indicative of the level of commitment the service requires and isreturned, along with the bounty, once the service is successfully completed. As partof determining a UWR, services must also define a proof for verifying its fulfillment.

Alternatively, service providers can set a subscription fee for dedicating their resourcesto a service. These dedicated providers are known as supernodes and their stakes areescrowed in a smart contract until the subscription ends. Because supernodes tendto be more reliable than their non-dedicated counterparts, they can control marketdynamics based on how they set their subscription fees. Supernodes in Bogotá mightcharge more than supernodes in Los Angeles due to higher hardware and internetcosts.

Network Effects

Introduction of new content to the network involves the invocation of DCDNsupernodes—Ara nodes dedicated to content redundancy and availability. Assumingthe bounty for a single file is constant, the effect of the marginal increase in fileavailability from a single peer sharing that file on the potential reward earnings fromthat file for any peer in DCDN (and all currency-based P2P file-sharing systemswith fixed incentives [2]) can be modeled using a submodular set function, which canbe intuitively thought of as a function describing diminishing returns. Due to thisproperty, DCDN supernodes can employ a subscription bounty model to counteractthe increasing opportunity cost of hosting content as its availability increases.

Content publishers can invoke and subscribe to as many or as few supernodes ingeographic locations of their choosing on a per content basis. Publishers, then, havefull freedom in deciding the extent to which they want their content to be readilyavailable globally. This enables support for the large media distributor who wantsto invoke all available supernodes worldwide to support a global audience, while alsoenabling support for the indie content creator who identifies their primary audienceas predominantly European and decides to prioritize European supernodes. Contentpublishers also determine the reward allocation for each content download. Thus,

15

there exists an optimal reward allocation and supernode distribution to achieve thedesired level of participation (i.e., file availability).

References

[1] Enterprise Strategy Group (2015, June), Price Comparison: GoogleCloud Platform vs. Amazon Web Services, https://cloud.google.com/files/esg-whitepaper.pdf

[2] M. Salek, S. Shayandeh, and D. Kempe, You Share, I Share: Net-work Effects and Economic Incentives in P2P File-Sharing Systemshttps://arxiv.org/pdf/1107.5559.pdf

16

II. DCDN Cost Analysis by Lester Kim

Introduction

In order to calculate the streaming costs for a potential partner, we need to knowhow much data B (in bytes) they need to deliver to consumers per unit of time T(in seconds). Let us have N groups of uploader nodes where N ∈ N. ∀n ∈ {1, ..., N},the average bandwidth of group n is bn (in bytes/second per node). Let qn be thequantity of group n nodes. Let b = [b1...bN ]ᵀ and q = [q1...qN ]ᵀ. Thus, the amountof bytes delivered per second constrained by B

T is

g(q) = b · q = B

T. (1)

We want to find the optimal q∗ to minimize the cost of distribution C(q). Ifp = [p1...pN ]ᵀ where pn is the price per node for group n, we have

C(q) = p · q. (2)

Uploader’s Profit Maximization

To determine p, let us visit the behavior of a profit maximizing firm. Let f be theproduction function with energy input E (in kWh) and output q (in nodes). Wemodel this production function as

f(E) = AEα (3)

where A is the factor of production (nodes/kWhα) and α ∈ [0, 1] is the elasticity ofproduction (percent increase in output over percent increase in input) [1].

Let P (in kWh/s) be the power increase when a node starts uploading data. Thisincludes sending data with its network interface controller (NIC) but can also includethe machine turning on (from either being off or in standby mode). If each nodehas power P , then for some E, a single node can run for E

P seconds. However, givena time constraint T to complete the work, there must be E

PT nodes. Thus,

A = D

(PT )α (4)

where D is the total factor productivity [2] (in nodes).

Let p be the price of a node and pE the price of energy (per kWh). The firm’s profitfunction π is

π(q, E) = pq − pEE. (5)

17

Bandwidth cost is ignored because it is a fixed cost in the short-term when lookingat seconds as opposed to months1.

We want to maximize the firm’s profit given an output requirement at least q;

maxq,E

π(q, E) s.t. f(E) ≥ q. (6)

To solve this, we take our Lagrangian to be

L(q, E, λ) = pq − pEE − λ(AEα − q). (7)

Taking partial derivatives and setting them to zero give

∂L∂q

= p+ λ = 0 (8)

∂L∂E

= −pE − λAαEα−1 = 0 (9)

∂L∂λ

= q −AEα = 0. (10)

Solving these first-order conditions gives

q∗ =(αA

1α p

pE

) α1−α

. (11)

Rewriting this (dropping the asterisk from q) gives

p = pEα

(q1−α

A

) 1α

. (12)

This formula lets the creator know what p should be to get the desired number ofnodes.

From (12), we find that the optimal revenue in terms of q is

pq = pEα

( qA

) 1α

. (13)

1Even with bandwidth included, its cost per second has the same order of magnitude as that of energy. InNYC, 50 MBps costs $3 x 10−5/second [3].

18

Distributor’s Cost Minimization

Since the revenue for the firm is spending for the consumer (the creator), we canwrite (2) as

C(q) = pEα

N∑n=1

( qnAn

) 1α

. (14)

The creator’s cost minimization problem is

minq

C(q) s.t. g(q) ≥ B

T. (15)

The Lagrangian is

L(q, λ) = C(q)− λ(g(q)− B

T). (16)

The first-order conditions are

∂L∂q = ∂C

∂q − λ∂g

∂q = 0 (17)

∂L∂λ

= B

T− g(q) = 0. (18)

From (14), (4), and (1),

∂C

∂q = pETPα2 ◦ (q◦(1−α) �D)◦ 1

α (19)

∂g

∂q = b (20)

where (D,P) = ([D1...DN ]ᵀ, [P1...PN ]ᵀ). "◦", "◦", "�" are the Hadamard (entrywise)product, power, and division, respectively [4]. So ∀m,n ∈ {1, ..., N},

Pαmq1−αm

bαmDm= Pαn q

1−αn

bαnDn. (21)

Thus,

bmqm =(bmDmP

αn

PαmbnDn

) 11−α

bnqn. (22)

19

Combining (22) with (18) gives

q∗ = Bb◦−1

Tκ◦ (b ◦D�P◦α)◦

11−α (23)

C∗ = pET

α

(B

Tκ1−α

) 1α

(24)

where

κ ≡N∑m=1

(bmDm

Pαm

) 11−α

. (25)

Case: α = 1

When α = 1, (23) and (24) become

q∗n ={ B|Υ|Tbn n ∈ Υ0 n /∈ Υ (26)

C∗ = pEBPnbnDn

any n ∈ Υ (27)

where

Υ ≡{n ∈ {1, ..., N}

∣∣∣ n = arg max1≤m≤N

bmDm

Pm

}. (28)

∀n ∈ Υ, each node in group n would deliver B|Υ|q∗

n(= bnT ) of data and receive at

least pEPnTDn

in compensation. However, there are multiple solutions to q∗. Forexample, for any n ∈ Υ, group n can take on all the work by employing B

Tbnnodes.

20

Example

Let’s work out an example in NYC where

α = 1 (29)B = 1 GB (30)N = 2 (31)pE = $0.2321/kWh [5] (32)T = 1 s (33)

b =[100 MB/s1 MB/s

][6] (34)

D =[1 node1 node

](35)

P =[200 W2 W

][7][8] (36)

to find examples of q∗ and C∗ for a creator. Then,

q∗ =[

5 nodes500 nodes

](37)

C∗ ≈ $1.29 x 10−4. (38)

This is 155 to 659 times (99.35% - 99.85%) cheaper than AWS Cloudfront’s on-demand pricing ($0.020/GB - $0.085/GB) [9]. Each group 1 node would handle 100MB whereas each group 2 node would handle 1 MB. Each node in group 1 and 2would need more than $1.29 x 10−5 and $1.29 x 10−7, respectively.

To put this in perspective, assume Netflix is a potential partner. In 2017, Netflixaveraged more than 140 million hours of content watched per day [10]. On average,a Netflix video is one GB/hour [11]. On the Ara platform, the 51.1-exabyte [12]annual spend would only be $6.6 million/year ($0.2106/second). If we estimateNetflix’s streaming cost to be $0.03/GB [13], we get $1.5 billion/year ($46.61/second).Using the Ara network would nearly quadruple Netflix’s 2017 net income of $558.9million [14]. (Manhattan has 1.66 million people [15] with 287,008 Netflix users2

streaming 321.45 TB/day, 3.72 GB/s. That requires 3,721 group 2 nodes at a rateof $41.47/day).

2At the end of 2018 Q1, Netflix had 56.71 million U.S. subscribers and 125 million worldwide [16]. There are328 million people in the U.S. [17], so proportionally, there are 287,008 Netflix subscribers in Manhattan.

21

References

[1] Wikipedia (2018, April 22), Cobb–Douglas production function,https://en.wikipedia.org/wiki/Cobb%E2%80%93Douglas_production_function

[2] Wikipedia (2018, June 5), Total factor productivity,https://en.wikipedia.org/wiki/Total_factor_productivity

[3] Spectrum (2017, December 29), Broadband Label Disclosure, p.2,https://www.spectrum.com/content/dam/spectrum/residential/en/pdfs/policies/Broadband_Label_Disclosure_Charter_122917.pdf

[4] Wikipedia (2018, March 10), Hadamard product (matrices),https://en.wikipedia.org/wiki/Hadamard_product_(matrices)

[5] Electricity Local (2018, June 19), https://www.electricitylocal.com/states/new-york/new-york

[6] Wikipedia (2018, June 19), Bandwidth (computing),https://en.wikipedia.org/wiki/Bandwidth_(computing)

[7] Energuide.be (2018, June 19), How much power does a computer use? Andhow much CO2 does that represent?, https://www.energuide.be/en/questions-answers/how-much-power-does-a-computer-use-and-how-much-co2-does-that-represent/54/

[8] R. Sohan, A. Rice, A. W. Moore, and K. Mansley, “Characterizing 10 GbpsNetwork Interface Energy Consumption,” The 35th Annual IEEE Conferenceon Local Computer Networks (LCN) Short Papers, University of Cambridge,Computer Laboratory, July 2010, https://www.cl.cam.ac.uk/ acr31/pubs/sohan-10gbpower.pdf.

[9] Amazon Web Services Pricing (2018, June 19), Amazon Cloudfront Pricing,https://aws.amazon.com/cloudfront/pricing

[10] L. Matney (2017, December 11), "Netflix users collectivelywatched 1 billion hours of content per week in 2017," Techcrunch,https://techcrunch.com/2017/12/11/netflix-users-collectively-watched-1-billion-hours-of-content-per-week-in-2017

[11] K. Hubby (2017, May 23), "The surprising amount of data Netflix uses," TheDaily Dot, https://www.dailydot.com/debug/how-much-data-netflix-use/

[12] Wikipedia (2018, June 20), Exabyte, https://en.wikipedia.org/wiki/Exabyte

[13] D. Rayburn (2009, July), "Stream This!: Netflix’sStreaming Costs," Streaming Media (June/July 2009),http://www.streamingmedia.com/Articles/Editorial/Featured-Articles/Stream-This!-Netflixs-Streaming-Costs-65503.aspx

[14] Netflix (2018, January 1), p.40, https://ir.netflix.com/static-files/20c3228d-bf1f-4956-a169-c8b76911ecd5

[15] Wikipedia (2018, July 5), Manhattan, https://en.wikipedia.org/wiki/Manhattan

22

[16] Statistica (2018, July 5), Number of Netflix streaming subscribers inthe United States from 3rd quarter 2011 to 1st quarter 2018 (in mil-lions), https://www.statista.com/statistics/250937/quarterly-number-of-netflix-streaming-subscribers-in-the-us/

[17] United States Census Bureau (2018, July 5), U.S. and World Population Clock,https://www.census.gov/popclock/

23

III. Expected Ara Rewards Analysis by Lester Kim

Network Model

Let us represent the Ara network of nodes as a complete graph [1] G = (V,E) where V containsN vertices with each vertex representing a node and E containing N(N−1)

2 edges with each edgerepresenting a communication channel between two nodes. Let C be some collection of content(in our case, a set of digital entertainment files) that all the nodes want to have, and let thesubset S ⊆ V contain all the nodes that have C (i.e. S = {v ∈ V : C ∈ v}).

Let time t ∈ N. At t = 0, |S| = 1, so there is only one v0 ∈ V that has content C; thus, thereis only one vertex that can deliver a copy of C to other nodes in the network. Assume that allother N − 1 nodes want C, and v0 has enough bandwidth to deliver C to only one node. Fromt = 0 to t = 1, |S| increases from 1 to 2. In general, at time t,

|S| ={

2t 0 ≤ t < log2N

N t ≥ log2N.(39)

Note that |S| = N starting at t = dlog2Ne.

∀s ∈ S, s will deliver C to some v ∈ V \ S only if v pays s an amount p. Let M be the network’stotal budget for entertainment delivery. Dividing this evenly by N nodes gives p = M/N .

At t = 0, the sole v0 ∈ S receives p from some v1 ∈ V \ S. Then, at t = 1, v0, v1 ∈ S eachreceives p from some v2, v3 ∈ V \ S. At any t < blog2Nc, each of the |S| = 2t nodes in Sreceives p from 2t nodes in V \ S. At t = blog2Nc, |S| > N

2 , so there are more suppliers thandemanders of C. When that occurs, N − 2t nodes from S are randomly selected to deliver C. Att = dlog2Ne, S = V .

In this model, v0 earns at least

Mblog2NcN

; (40)

v1 earns at least M(blog2 Nc−1)N ; vk earns at least

M(blog2Nc − dlog2 (k + 1)e)N

. (41)

The greatest k such that vk earns at least MN is when

blog2Nc − dlog2 (k + 1)e ≥ 1 (42)

which implies

log2 (k + 1) ≤ log2N

2 . (43)

Thus, the maximum value of k to earn at least MN is k = bN2 c − 1. On average, each earns

24

M − MN

2dlog2 Ne−1 =M(1− 1

N )2dlog2 Ne−1 . (44)

The numerator is M − MN to exclude v0’s entertainment budget since it had C at t = 0. The

denominator is 2dlog2 Ne−1 because at t = dlog2Ne − 1, |S| = 2dlog2 Ne−1, and at that point, Sconsists of all the nodes that have the possibility of earning rewards throughout this process.This means that there are N − 2dlog2 Ne−1 nodes that will not be able to earn rewards.

Example

Approximately 80% of Americans have computers with Internet access [2]. Since there are 327MAmericans living in the US [3], there are (0.8)(327M) = 261.6M Americans with devices connectedto the Internet. Assuming each has one device, let N = 261.6M. The annual US entertainmentconsumption is $734B [4] [5]. Let’s assume most of this expenditure for future years will be digital,but let’s only include the budget of the 80% of Americans who have Internet access such that thespending among them is (0.8)(734B) = $587B. Let 10% of the spending be for covering distributioncosts. Then, M = (0.1)($587B) = $58.7B. Then, p = M/N = $58.7B/261.6M = $224.39. By(44), the average annual earnings is $437.35 per node. By (40), the most v0 can earn is $6282.87.Thus, this example illustrates that the initial peers who share content will earn the most rewards.

References

[1] Wikipedia (2018, June 19), Complete graph, https://en.wikipedia.org/wiki/Complete_graph

[2] C. Ryan and J. M. Lewis, “Computer and Internet Use in the United States:2015,” American Community Survey Reports U.S. Census Bureau, September 2017https://www.census.gov/content/dam/Census/library/publications/2017/acs/acs-37.pdf

[3] World Population Review (2018, June 18) United States Population 2018http://worldpopulationreview.com/countries/united-states-population/

[4] SelectUSA (2018, August 22), MEDIA AND ENTERTAINMENT SPOTLIGHT,https://www.selectusa.gov/media-entertainment-industry-united-states

[5] Bureau of Labor Statistics (2017, August 29), CONSUMER EXPENDITURES–2016https://www.bls.gov/news.release/cesan.nr0.htm

25